parse.py 167 KB
Newer Older
1
# parse.py - parser combinators for DHParser
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
#
# Copyright 2016  by Eckhart Arnold (arnold@badw.de)
#                 Bavarian Academy of Sciences an Humanities (badw.de)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.  See the License for the specific language governing
# permissions and limitations under the License.
Eckhart Arnold's avatar
Eckhart Arnold committed
17

18
19

"""
20
Module ``parse`` contains the python classes and functions for
21
22
DHParser's packrat-parser. It's central class is the
``Grammar``-class, which is the base class for any concrete
23
24
25
26
27
28
29
30
31
Grammar. Grammar-objects are callable and parsing is done by
calling a Grammar-object with a source text as argument.

The different parsing functions are callable descendants of class
``Parser``. Usually, they are organized in a tree and defined
within the namespace of a grammar-class. See ``ebnf.EBNFGrammar``
for an example.
"""

32

33
from collections import defaultdict
34
import copy
35
from typing import Callable, cast, List, Tuple, Set, AbstractSet, Dict, \
36
    DefaultDict, Sequence, Union, Optional, Iterator, Hashable, NamedTuple
37

eckhart's avatar
eckhart committed
38
from DHParser.configuration import get_config_value
Eckhart Arnold's avatar
Eckhart Arnold committed
39
from DHParser.error import Error, ErrorCode, MANDATORY_CONTINUATION, \
40
    UNDEFINED_RETRIEVE, PARSER_LOOKAHEAD_FAILURE_ONLY, \
eckhart's avatar
eckhart committed
41
    PARSER_LOOKAHEAD_MATCH_ONLY, PARSER_STOPPED_BEFORE_END, PARSER_NEVER_TOUCHES_DOCUMENT, \
42
43
44
    MALFORMED_ERROR_STRING, MANDATORY_CONTINUATION_AT_EOF, DUPLICATE_PARSERS_IN_ALTERNATIVE, \
    CAPTURE_WITHOUT_PARSERNAME, CAPTURE_DROPPED_CONTENT_WARNING, LOOKAHEAD_WITH_OPTIONAL_PARSER, \
    BADLY_NESTED_OPTIONAL_PARSER, BAD_ORDER_OF_ALTERNATIVES, BAD_MANDATORY_SETUP, \
45
    OPTIONAL_REDUNDANTLY_NESTED_WARNING, CAPTURE_STACK_NOT_EMPTY, BAD_REPETITION_COUNT, \
46
    AUTORETRIEVED_SYMBOL_NOT_CLEARED, RECURSION_DEPTH_LIMIT_HIT, SourceMapFunc
eckhart's avatar
eckhart committed
47
from DHParser.log import CallItem, HistoryRecord
48
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
49
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
eckhart's avatar
eckhart committed
50
from DHParser.syntaxtree import ChildrenType, Node, RootNode, WHITESPACE_PTYPE, \
51
    TOKEN_PTYPE, ZOMBIE_TAG, EMPTY_NODE, ResultType
52
from DHParser.toolkit import sane_parser_name, escape_ctrl_chars, re, cython, \
53
    abbreviate_middle, RX_NEVER_MATCH, RxPatternType, linebreaks, line_col, identity
54

55

56
__all__ = ('ParserError',
eckhart's avatar
eckhart committed
57
58
59
           'ApplyFunc',
           'FlagFunc',
           'ParseFunc',
60
           'Parser',
61
           'AnalysisError',
62
           'GrammarError',
63
           'Grammar',
64
65
           'Always',
           'Never',
eckhart's avatar
eckhart committed
66
           'AnyChar',
67
           'PreprocessorToken',
68
69
           'Text',
           'DropText',
70
           'RegExp',
71
           'update_scanner',
72
73
           'RE',
           'TKN',
74
           'Whitespace',
75
           'DropRegExp',
76
           'mixin_comment',
di68kap's avatar
di68kap committed
77
           'mixin_nonempty',
di68kap's avatar
di68kap committed
78
           'CombinedParser',
79
           'TreeReduction',
eckhart's avatar
eckhart committed
80
81
           'UnaryParser',
           'NaryParser',
82
           'Drop',
83
84
85
86
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
87
           'NO_MANDATORY',
Eckhart Arnold's avatar
Eckhart Arnold committed
88
           'MandatoryNary',
89
90
           'Series',
           'Alternative',
91
           'INFINITE',
92
           'Counted',
93
           'Interleave',
eckhart's avatar
eckhart committed
94
           'Required',
95
96
97
98
99
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
100
101
           'optional_last_value',
           'matching_bracket',
102
103
104
           'Capture',
           'Retrieve',
           'Pop',
105
           'Forward')
106
107
108
109


########################################################################
#
110
# ParserError class
111
112
113
114
#
########################################################################


Eckhart Arnold's avatar
Eckhart Arnold committed
115
116
117
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
118
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
119
    If a reentry-rule has been configured for the parser where the error
120
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
121
122

    Currently, the only case when a `ParserError` is thrown (and not some
123
    different kind of error like `UnknownParserError`) is when a `Series`-
eckhart's avatar
eckhart committed
124
    or `Interleave`-parser detects a missing mandatory element.
Eckhart Arnold's avatar
Eckhart Arnold committed
125
    """
126
127
128
129
130
131
    def __init__(self,
                 parser: 'Parser',
                 node: Node,
                 rest: StringView,
                 error: Error, *,
                 first_throw: bool):
eckhart's avatar
eckhart committed
132
        assert node is not None
133
134
135
136
137
138
        self.parser = parser  # type: 'Parser'
        self.node = node      # type: Node
        self.rest = rest      # type: StringView
        self.error = error    # type: Error
        self.first_throw = first_throw   # type:
        self.attributes_locked = frozenset({'parser', 'node', 'rest', 'error', 'first_throw'})
eckhart's avatar
eckhart committed
139
        self.frozen_callstack = tuple()  # type: Tuple[CallItem, ...]  # tag_name, location
Eckhart Arnold's avatar
Eckhart Arnold committed
140

141
142
143
144
145
146
147
148
149
    def __setattr__(self, name, value):
        if name == "attributes_locked":
            self.__dict__[name] = value
        elif "attributes_locked" not in self.__dict__ \
                or name not in self.__dict__['attributes_locked']:
            self.__dict__[name] = value
        else:
            raise TypeError('Attribute %s of ParserError-object must not be reassigned!' % name)

150
    def __str__(self):
151
152
        return "%i: %s    %s (%s)" \
               % (self.node.pos, str(self.rest[:25]), repr(self.node), str(self.error))
153

Eckhart Arnold's avatar
Eckhart Arnold committed
154

155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
    def new_PE(self, **kwargs):
        """Returns a new ParserError object with the same attribute values
        as `self`, except those that are reassigned in `**kwargs`.

        >>> pe = ParserError(Parser(), Node('test', ""), StringView(""), Error("", 0), first_throw=True)
        >>> pe_derived = pe.new_PE(first_throw = False)
        >>> pe.first_throw
        True
        >>> pe_derived.first_throw
        False
        """
        args = { "parser": self.parser,
                 "node": self.node,
                 "rest": self.rest,
                 "error": self.error,
                 "first_throw": self.first_throw }
        assert len(kwargs.keys() - args.keys()) == 0, str(kwargs.keys() - args.keys())
        args.update(kwargs)
        pe = ParserError(**args)
        pe.frozen_callstack = self.frozen_callstack
        return pe


178
179
180
PatternMatchType = Union[RxPatternType, str, Callable]
ErrorMessagesType = List[Tuple[PatternMatchType, str]]
ResumeList = List[PatternMatchType]  # list of strings or regular expressions
181
182
ReentryPointAlgorithm = Callable[[StringView, int, int], Tuple[int, int]]
# (text, start point, end point) => (reentry point, match length)
183
# A return value of (-1, x) means that no reentry point before the end of the document was found
Eckhart Arnold's avatar
Eckhart Arnold committed
184

185

186
@cython.returns(cython.int)
eckhart's avatar
eckhart committed
187
@cython.locals(upper_limit=cython.int, closest_match=cython.int, pos=cython.int)
188
189
190
def reentry_point(rest: StringView,
                  rules: ResumeList,
                  comment_regex,
191
                  search_window: int = -1) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
192
193
    """
    Finds the point where parsing should resume after a ParserError has been caught.
194
    The algorithm makes sure that this reentry-point does not lie inside a comment.
195
196
197
    The re-entry point is always the point after the end of the match of the regular
    expression defining the re-entry point. (Use look ahead, if you wand to define
    the re-entry point by what follows rather than by what text precedes the point.)
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223

    REMARK: The algorithm assumes that any stretch of the document that matches
    `comment_regex` is actually a comment. It is possible to define grammars,
    where the use of comments is restricted to certain areas and that allow to
    use constructs that look like comments (i.e. will be matched by `comment_regex`)
    but are none in other areas. For example::

            my_string = "# This is not a comment"; foo()  # This is a comment
            bar()

    Here the reentry-algorithm would overlook `foo()` and jump directly to `bar()`.
    However, since the reentry-algorithm only needs to be good enough to do its
    work, this seems acceptable.

    :param rest:  The rest of the parsed text or, in other words, the point where
        a ParserError was thrown.
    :param rules: A list of strings, regular expressions or search functions.
        The rest of the text is searched for each of these. The closest match
        is the point where parsing will be resumed.
    :param comment_regex: A regular expression object that matches comments.
    :param search_window: The maximum size of the search window for finding the
        reentry-point. A value smaller than zero means that the complete remaining
        text will be searched. A value of zero effectively turns of resuming after
        error.
    :return: The integer index of the closest reentry point or -1 if no
        reentry-point was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
224
    """
225
    upper_limit = len(rest) + 1
226
    closest_match = upper_limit
227
228
229
    comments = None  # type: Optional[Iterator]
    if search_window < 0:
        search_window = len(rest)
230

eckhart's avatar
eckhart committed
231
    @cython.locals(a=cython.int, b=cython.int)
232
    def next_comment() -> Tuple[int, int]:
233
234
235
236
        """Returns the [start, end[ intervall of the next comment in the text.
        The comment-iterator start at the beginning of the `rest` of the
        document and is reset for each search rule.
        """
237
238
239
240
241
242
243
244
245
246
        nonlocal rest, comments
        if comments:
            try:
                m = next(comments)
                a, b = m.span()
                return rest.index(a), rest.index(b)
            except StopIteration:
                comments = None
        return -1, -2

247
248
    @cython.locals(start=cython.int)
    def str_search(s, start: int = 0) -> Tuple[int, int]:
249
250
251
252
253
        """Returns the starting position of the next occurrence of `s` in
        the `rest` of the document beginning with `start` and the length
        of the match, which in this case is always the length of `s` itself.
        If their is no match, the returned starting position will be -1.
        """
254
255
        nonlocal rest
        return rest.find(s, start, start + search_window), len(s)
256

eckhart's avatar
eckhart committed
257
    @cython.locals(start=cython.int, end=cython.int)
258
    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
259
260
261
262
263
        """Returns the staring position and the length of the next match of
        the regular expression `rx` in the `rest` of the document, starting
        with `start`.
        If their is no match, the returned starting position will be -1.
        """
264
        nonlocal rest
265
        m = rest.search(rx, start, start + search_window)
266
        if m:
267
268
            begin, end = m.span()
            return rest.index(begin), end - begin
269
270
        return -1, 0

271
272
273
274
    def algorithm_search(func: ReentryPointAlgorithm, start: int = 0):
        """Returns the next match as a tuple of position and length that
        the reentry-point-search-function `func` yields.
        """
275
        nonlocal rest
276
        return func(rest, start, start + search_window)
277

278
    @cython.returns(cython.int)
eckhart's avatar
eckhart committed
279
    @cython.locals(a=cython.int, b=cython.int, k=cython.int, length=cython.int)
280
    def entry_point(search_func, search_rule) -> int:
281
282
283
        """Returns the next reentry-point outside a comment that `search_func`
        yields. If no reentry point is found, the first position after the
        end of the text ("upper limit") is returned."""
284
285
        a, b = next_comment()
        k, length = search_func(search_rule)
eckhart's avatar
eckhart committed
286
        while a < b <= k + length:
287
            a, b = next_comment()
288
289
290
        # find next as long as start or end point of resume regex are inside a comment
        while (a < k < b) or (a < k + length < b):
            k, length = search_func(search_rule, b)
291
292
            while a < b <= k:
                a, b = next_comment()
293
        return k + length if k >= 0 else upper_limit
294

295
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
296
    for rule in rules:
297
        comments = rest.finditer(comment_regex)
298
299
300
301
302
303
304
        if callable(rule):
            search_func = algorithm_search
        elif isinstance(rule, str):
            search_func = str_search
        else:
            search_func = rx_search
        pos = entry_point(search_func, rule)
305
        closest_match = min(pos, closest_match)
306

307
    # in case no rule matched return -1
308
309
310
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
311
312


313
314
315
316
317
318
319
########################################################################
#
# Parser base class
#
########################################################################


320
ApplyFunc = Callable[[List['Parser']], Optional[bool]]
321
# The return value of `True` stops any further application
eckhart's avatar
eckhart committed
322
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]
eckhart's avatar
eckhart committed
323
ParseFunc = Callable[[StringView], Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
324
325


326
class Parser:
327
328
329
330
331
332
333
334
335
336
337
338
339
340
    """
    (Abstract) Base class for Parser combinator parsers. Any parser
    object that is actually used for parsing (i.e. no mock parsers)
    should should be derived from this class.

    Since parsers can contain other parsers (see classes UnaryOperator
    and NaryOperator) they form a cyclical directed graph. A root
    parser is a parser from which all other parsers can be reached.
    Usually, there is one root parser which serves as the starting
    point of the parsing process. When speaking of "the root parser"
    it is this root parser object that is meant.

    There are two different types of parsers:

di68kap's avatar
di68kap committed
341
    1. *Named parsers* for which a name is set in field `parser.pname`.
342
343
344
       The results produced by these parsers can later be retrieved in
       the AST by the parser name.

345
346
    2. *Disposable parsers* where the name-field just contains the empty
       string. AST-transformation of disposable parsers can be hooked
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
       only to their class name, and not to the individual parser.

    Parser objects are callable and parsing is done by calling a parser
    object with the text to parse.

    If the parser matches it returns a tuple consisting of a node
    representing the root of the concrete syntax tree resulting from the
    match as well as the substring `text[i:]` where i is the length of
    matched text (which can be zero in the case of parsers like
    `ZeroOrMore` or `Option`). If `i > 0` then the parser has "moved
    forward".

    If the parser does not match it returns `(None, text). **Note** that
    this is not the same as an empty match `("", text)`. Any empty match
    can for example be returned by the `ZeroOrMore`-parser in case the
    contained parser is repeated zero times.

    Attributes and Properties:
365

366
        pname:  The parser's name.
367

368
369
        disposable: A property indicating that the parser returns
                anonymous nodes. For performance
370
                reasons this is implemented as an object variable rather
371
                than a property. This property should always be equal to
372
                `self.tag_name[0] == ":"`.
373

374
375
376
377
        drop_content: A property (for performance reasons implemented as
                simple field) that, if set, induces the parser not to return
                the parsed content or sub-tree if it has matched but the
                dummy `EMPTY_NODE`. In effect the parsed content will be
378
379
380
                dropped from the concrete syntax tree already. Only
                anonymous (or pseudo-anonymous) parsers are allowed to
                drop content.
381

382
        tag_name: The tag_name for the nodes that are created by
383
                the parser. If the parser is named, this is the same as
di68kap's avatar
di68kap committed
384
385
                `pname`, otherwise it is the name of the parser's type
                prefixed with a colon ":".
386

387
388
389
390
391
392
393
        eq_class: A unique number for the class of functionally
                equivalent parsers that this parser belongs to.
                (This serves the purpose of optimizing memoization,
                by tying memoization dictionaries to the classes
                of functionally equivalent parsers, rather than to
                the individual parsers themselves.)

394
395
396
397
398
399
400
401
402
403
        visited:  Mapping of places this parser has already been to
                during the current parsing process onto the results the
                parser returned at the respective place. This dictionary
                is used to implement memoizing.

        cycle_detection:  The apply()-method uses this variable to make
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.

404
405
406
407
        proxied: The original `_parse()`-method is stored here, if a
                proxy (e.g. a tracing debugger) is installed via the
                `set_proxy()`-method.

408
        _grammar:  A reference to the Grammar object to which the parser
409
                is attached.
410
411
412
413
414

        _symbol:  The name of the closest named parser to which this
                parser is connected in a grammar. If pname is not the
                empty string, this will become the same as pname, when
                the property `symbol` is read for the first time.
415
416
    """

417
    def __init__(self) -> None:
418
        # assert isinstance(name, str), str(name)
419
        self.pname = ''               # type: str
420
        self.disposable = True        # type: bool
421
        self.drop_content = False     # type: bool
422
        self.tag_name = self.ptype    # type: str
423
        self.eq_class = id(self)      # type: int
424
        self.cycle_detection = set()  # type: Set[ApplyFunc]
425
        # this indirection is required for Cython-compatibility
426
        self._parse_proxy = self._parse  # type: ParseFunc
eckhart's avatar
eckhart committed
427
        # self.proxied = None           # type: Optional[ParseFunc]
428
        try:
429
            self._grammar = get_grammar_placeholder()  # type: Grammar
430
        except NameError:
di68kap's avatar
di68kap committed
431
            pass                      # ensures Cython-compatibility
432
        self._symbol = ''             # type: str
433
434
435
        self.reset()

    def __deepcopy__(self, memo):
436
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
437
        object, parsers will be deep-copied to the Grammar object. If a
438
        derived parser-class changes the signature of the `__init__`-constructor,
439
440
441
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
442
        duplicate = self.__class__()
443
        copy_parser_base_attrs(self, duplicate)
444
        return duplicate
445

446
    def __repr__(self):
di68kap's avatar
di68kap committed
447
        return self.pname + self.ptype
448
449

    def __str__(self):
di68kap's avatar
di68kap committed
450
        return self.pname + (' = ' if self.pname else '') + repr(self)
451

452
453
454
455
456
457
    @property
    def ptype(self) -> str:
        """Returns a type name for the parser. By default this is the name of
        the parser class with an added leading colon ':'. """
        return ':' + self.__class__.__name__

458
459
460
461
462
    @property
    def symbol(self) -> str:
        """Returns the symbol with which the parser is associated in a grammar.
        This is the closest parser with a pname that contains this parser."""
        if not self._symbol:
eckhart's avatar
eckhart committed
463
            try:
464
                self._symbol = self.grammar.associated_symbol__(self).pname
eckhart's avatar
eckhart committed
465
466
467
468
            except AttributeError:
                # return an empty string, if parser is not connected to grammar,
                # but be sure not to save the empty string in self._symbol
                return ''
469
470
        return self._symbol

471
472
473
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
474
        return self.pname if self.pname else self.__repr__()
475

476
477
478
479
    def reset(self):
        """Initializes or resets any parser variables. If overwritten,
        the `reset()`-method of the parent class must be called from the
        `reset()`-method of the derived class."""
480
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
481

eckhart's avatar
eckhart committed
482
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
483
484
485
    def __call__(self: 'Parser', text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given text. This is a wrapper method that adds
        the business intelligence that is common to all parsers. The actual parsing is
486
487
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
488
        """
489
        grammar = self._grammar
di68kap's avatar
di68kap committed
490
        location = grammar.document_length__ - text._len  # faster then len(text)?
491
492

        try:
493
494
            # rollback variable changing operation if parser backtracks to a position
            # before or at the location where the variable changing operation occurred
495
            if location <= grammar.last_rb__loc__:
496
497
                grammar.rollback_to__(location)

498
            # if location has already been visited by the current parser, return saved result
499
            visited = self.visited  # using local variable for better performance
di68kap's avatar
di68kap committed
500
            if location in visited:
501
                # no history recording in case of memoized results!
di68kap's avatar
di68kap committed
502
                return visited[location]
503

504
505
            memoization_state = grammar.suspend_memoization__
            grammar.suspend_memoization__ = False
506

507
            # now, the actual parser call!
508
            try:
Eckhart Arnold's avatar
Eckhart Arnold committed
509
                node, rest = self._parse_proxy(text)
510
            except ParserError as pe:
511
                # catching up with parsing after an error occurred
512
                gap = len(text) - len(pe.rest)
513
                rules = grammar.resume_rules__.get(
514
                    self.pname or grammar.associated_symbol__(self).pname, [])
515
                rest = pe.rest[len(pe.node):]
516
517
                i = reentry_point(rest, rules, grammar.comment_rx__,
                                  grammar.reentry_search_window__)
518
                if i >= 0 or self == grammar.start_parser__:
eckhart's avatar
eckhart committed
519
520
                    # either a reentry point was found or the
                    # error has fallen through to the first level
521
                    assert pe.node._children or (not pe.node.result)
522
                    # apply reentry-rule or catch error at root-parser
eckhart's avatar
eckhart committed
523
                    if i < 0:  i = 0
524
                    try:
525
                        zombie = pe.node.pick_child(ZOMBIE_TAG)  # type: Optional[Node]
526
                    except (KeyError, ValueError):
527
528
529
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
eckhart's avatar
eckhart committed
530
                        tail = tuple()  # type: ChildrenType
531
532
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
eckhart's avatar
eckhart committed
533
                        # nd.attr['err'] = pe.error.message
534
                        tail = (nd,)
535
                    rest = rest[i:]
536
537
                    if pe.first_throw:
                        node = pe.node
538
                        node.result = node._children + tail
539
                    else:
540
541
                        node = Node(
                            self.tag_name,
di68kap's avatar
di68kap committed
542
543
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail) \
                            .with_pos(location)
eckhart's avatar
eckhart committed
544
                # if no re-entry point was found, do any of the following:
545
                elif pe.first_throw:
eckhart's avatar
eckhart committed
546
                    # just fall through
eckhart's avatar
eckhart committed
547
                    # TODO: Is this case still needed with module "trace"?
548
                    raise pe.new_PE(first_throw=False)
549
                elif grammar.tree__.errors[-1].code == MANDATORY_CONTINUATION_AT_EOF:
eckhart's avatar
eckhart committed
550
551
                    # try to create tree as faithful as possible
                    node = Node(self.tag_name, pe.node).with_pos(location)
552
                else:
eckhart's avatar
eckhart committed
553
                    # fall through but skip the gap
554
555
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
556
557
                    raise pe.new_PE(node=Node(self.tag_name, result).with_pos(location),
                                    rest=text, first_throw=False)
558

559
            if node is None:
di68kap's avatar
di68kap committed
560
561
562
                if location > grammar.ff_pos__:
                    grammar.ff_pos__ = location
                    grammar.ff_parser__ = self
563
            else:
564
                node._pos = location
565
566
567
            if not grammar.suspend_memoization__:
                visited[location] = (node, rest)
                grammar.suspend_memoization__ = memoization_state
568

569
        except RecursionError:
570
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
571
            node._pos = location
572
573
574
            error = Error("maximum recursion depth of parser reached; potentially due to too many "
                          "errors or left recursion!", location, RECURSION_DEPTH_LIMIT_HIT)
            grammar.tree__.add_error(node, error)
575
            grammar.most_recent_error__ = ParserError(self, node, text, error, first_throw=False)
576
577
578
            rest = EMPTY_STRING_VIEW

        return node, rest
579
580
581
582

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
di68kap's avatar
di68kap committed
583
584
        if isinstance(other, Series):
            return cast('Series', other).__radd__(self)
585
586
587
588
589
590
        return Series(self, other)

    def __or__(self, other: 'Parser') -> 'Alternative':
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
di68kap's avatar
di68kap committed
591
592
        if isinstance(other, Alternative):
            return cast('Alternative', other).__ror__(self)
593
594
        return Alternative(self, other)

595
    def __mul__(self, other: 'Parser') -> 'Interleave':
di68kap's avatar
di68kap committed
596
597
598
599
600
601
602
603
        """The * operator generates an interleave parser that applies
        the first parser and the second parser in any possible order
        until both match.
        """
        if isinstance(other, Interleave):
            return cast(Interleave, other).__rmul__(self)
        return Interleave(self, other)

604
605
606
607
608
609
    def _parse(self, text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given `text` and returns a node with
        the results or None as well as the text at the position right behind
        the matching string."""
        raise NotImplementedError

eckhart's avatar
eckhart committed
610
    def is_optional(self) -> Optional[bool]:
611
        """Returns `True`, if the parser can never fail, i.e. never yields
eckhart's avatar
eckhart committed
612
613
614
615
616
        `None`, instead of a node. Returns `False`, if the parser can fail.
        Returns `None` if it is not known whether the parser can fail.
        """
        return None

617
    def set_proxy(self, proxy: Optional[ParseFunc]):
618
619
620
        """Sets a proxy that replaces the _parse()-method. Call `set_proxy`
        with `None` to remove a previously set proxy. Typical use case is
        the installation of a tracing debugger. See module `trace`.
621
622
        """
        if proxy is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
623
            self._parse_proxy = self._parse
624
        else:
eckhart's avatar
eckhart committed
625
626
            if not isinstance(proxy, type(self._parse)):
                # assume that proxy is a function and bind it to self
627
628
                proxy = proxy.__get__(self, type(self))
            else:
eckhart's avatar
eckhart committed
629
                # if proxy is a method it must be a method of self
630
                assert proxy.__self__ == self
Eckhart Arnold's avatar
Eckhart Arnold committed
631
            self._parse_proxy = cast(ParseFunc, proxy)
632

633
    def name(self, pname: str) -> 'Parser':
634
635
636
        """Sets the parser name to `pname` and returns `self`."""
        self.pname = pname
        self.tag_name = pname or self.ptype
637
        self.disposable = not self.pname
638
639
        return self

640
    @property
eckhart's avatar
eckhart committed
641
    def grammar(self) -> 'Grammar':
642
        try:
643
            if not is_grammar_placeholder(self._grammar):
644
645
                return self._grammar
            else:
eckhart's avatar
eckhart committed
646
                raise ValueError('Grammar has not yet been set!')
647
648
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
649
650
651

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
652
        try:
653
            if is_grammar_placeholder(self._grammar):
654
655
656
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
eckhart's avatar
eckhart committed
657
658
                raise AssertionError("Parser has already been assigned"
                                     "to a different Grammar object!")
659
660
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
661
        except NameError:  # Cython: No access to _GRAMMAR_PLACEHOLDER, yet :-(
662
            self._grammar = grammar
663

eckhart's avatar
eckhart committed
664
    def sub_parsers(self) -> Tuple['Parser', ...]:
665
666
667
        """Returns the list of sub-parsers if there are any.
        Overridden by Unary, Nary and Forward.
        """
eckhart's avatar
eckhart committed
668
        return tuple()
669

670
    def _apply(self, func: ApplyFunc, context: List['Parser'], flip: FlagFunc) -> bool:
671
672
        """
        Applies function `func(parser)` recursively to this parser and all
673
674
        descendant parsers as long as `func()` returns `None` or `False`.
        Otherwise stops the further application of `func` and returns `True`.
eckhart's avatar
eckhart committed
675
676
677
678
679
680
681
682

        In order to break cycles, function `flip` is called, which should
        return `True`, if this parser has already been visited. If not, it
        flips the cycle detection flag and returns `False`.

        This is a protected function and should not called from outside
        class Parser or any of its descendants. The entry point for external
        calls is the method `apply()` without underscore!
683
        """
684
685
        if not flip(func, self.cycle_detection):
            if func(context + [self]):
686
687
688
                return True
            else:
                for parser in self.sub_parsers():
689
                    if parser._apply(func, context + [self], flip):
690
691
692
                        return True
                return False
        return False
693

694
    def apply(self, func: ApplyFunc) -> Optional[bool]:
eckhart's avatar
eckhart committed
695
696
        """
        Applies function `func(parser)` recursively to this parser and all
697
698
699
700
701
702
703
704
705
706
707
708
709
710
        descendant parsers as long as `func()` returns `None` or `False`.
        Traversal is pre-order. Stops the further application of `func` and
        returns `True` once `func` has returned `True`.

        If `func` has been applied to all descendant parsers without issuing
        a stop signal by returning `True`, `False` is returned.

        This use of the return value allows to use the `apply`-method both
        to issue tests on all descendant parsers (including self) which may be
        decided already after some parsers have been visited without any need
        to visit further parsers. At the same time `apply` can be used to simply
        `apply` a procedure to all descendant parsers (including self) without
        worrying about forgetting the return value of procedure, because a
        return value of `None` means "carry on".
eckhart's avatar
eckhart committed
711
        """
712
        def positive_flip(f: ApplyFunc, flagged: AbstractSet[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
713
714
715
716
717
718
719
720
721
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == True`
            as meaning that `f` has already been applied."""
            if f in flagged:
                return True
            else:
                flagged.add(f)
                return False

722
        def negative_flip(f: ApplyFunc, flagged: AbstractSet[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
723
724
725
726
727
728
729
730
731
732
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == False`
            as meaning that `f` has already been applied."""
            if f not in flagged:
                return True
            else:
                flagged.remove(f)
                return False

        if func in self.cycle_detection:
733
            return self._apply(func, [], negative_flip)
eckhart's avatar
eckhart committed
734
        else:
735
            return self._apply(func, [], positive_flip)
736

737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
    def _signature(self) -> Hashable:
        """This method should be implemented by any non-abstract descendant
        parser class. The implementation must make sure that all instances
        that have the same signature always yield the same parsing result
        for the same piece of text.

        It does not hurt, but only wastes an opportunity for optimization,
        if two functionally equivalent parsers provide different signatures.
        It is a serious mistake, though, if two functionally non-equivalent
        parsers have the same signature.

        By returning `id(self)` this mistake will become impossible, but
        it will turn signature-based memoization-optimizatiopn off for
        this parser.
        """
        return id(self)

    def signature(self) -> Hashable:
        """Returns a value that is identical for two different
        parser objects if they are functionally equivalent, i.e.
        yield the same return value for the same call parameters:

        >>> a = Text('[')
        >>> b = Text('[')
        >>> c = Text(']')
        >>> a is b
        False
        >>> a.signature() == b.signature()
        True
        >>> a.signature() == c.signature()
        False

        The purpose of parser-signatures is to optimize better
        memoization in cases of code repetition in the grammar.

        DO NOT OVERRIDE THIS METHOD. In order to implement a
        signature function, the protected method `_signature`
        should be overridden instead.
        """
        return self.pname if self.pname else self._signature()


779
    def static_error(self, msg: str, code: ErrorCode) -> 'AnalysisError':
780
        return AnalysisError(self.symbol, self, Error(msg, 0, code))
781

eckhart's avatar
eckhart committed
782
    def static_analysis(self) -> List['AnalysisError']:
783
784
        """Analyses the parser for logical errors after the grammar has been
        instantiated."""
eckhart's avatar
eckhart committed
785
        return []
eckhart's avatar
eckhart committed
786

787

788
def copy_parser_base_attrs(src: Parser, duplicate: Parser):
789
    """Duplicates all attributes of the Parser-class from `src` to `duplicate`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
790
    duplicate.pname = src.pname
791
    duplicate.disposable = src.disposable
Eckhart Arnold's avatar
Eckhart Arnold committed
792
793
    duplicate.drop_content = src.drop_content
    duplicate.tag_name = src.tag_name
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
    duplicate.eq_class = src.eq_class


def determine_eq_classes(root: Parser):
    """Sorts the parsers originating in root (imperfectly) into equivalence
    classes and assigns respective the class identifier to the `eq_class`-field
    of each parser."""
    eq_classes: Dict[Hashable, int] = {}

    def assign_eq_class(parser_stack: List[Parser]) -> bool:
        nonlocal eq_classes
        p = parser_stack[-1]
        signature = p.signature()
        p.eq_class = eq_classes.set_default(signature, id(p))
        return False
Eckhart Arnold's avatar
Eckhart Arnold committed
809

810
    root.apply(assign_eq_class)
Eckhart Arnold's avatar
Eckhart Arnold committed
811

di68kap's avatar
di68kap committed
812

eckhart's avatar
eckhart committed
813
814
def Drop(parser: Parser) -> Parser:
    """Returns the parser with the `parser.drop_content`-property set to `True`."""
815
    assert parser.disposable, "Parser must be anonymous to be allowed to drop ist content."
eckhart's avatar
eckhart committed
816
817
818
    if isinstance(parser, Forward):
        cast(Forward, parser).parser.drop_content = True
    parser.drop_content = True
Eckhart Arnold's avatar
Eckhart Arnold committed
819
    return parser
eckhart's avatar
eckhart committed
820
821


822
823
824
825
826
827
PARSER_PLACEHOLDER = None  # type: Optional[Parser]


def get_parser_placeholder() -> Parser:
    global PARSER_PLACEHOLDER
    if PARSER_PLACEHOLDER is None:
di68kap's avatar
di68kap committed
828
829
830
831
        PARSER_PLACEHOLDER = Parser.__new__(Parser)  # Parser()
        PARSER_PLACEHOLDER.pname = ''
        PARSER_PLACEHOLDER.disposable = False
        PARSER_PLACEHOLDER.drop_content = False
di68kap's avatar
di68kap committed
832
        PARSER_PLACEHOLDER.tag_name = ':PLACEHOLDER__'
833
        PARSER_PLACEHOLDER.eq_class = id(PARSER_PLACEHOLDER)
834
    return cast(Parser, PARSER_PLACEHOLDER)
835
836


eckhart's avatar
eckhart committed
837
838
839
840
841
def is_parser_placeholder(parser: Optional[Parser]) -> bool:
    """Returns True, if `parser` is `None` or merely a placeholder for a parser."""
    return not parser or parser.ptype == ":Parser"


eckhart's avatar
eckhart committed
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
# functions for analysing the parser tree/graph ###


def has_non_autocaptured_symbols(context: List[Parser]) -> Optional[bool]:
    """Returns True, if the context contains a Capture-Parser that is not
    shielded by a Retrieve-Parser. This is the case for captured symbols
    that are not "auto-captured" by a Retrieve-Parser.
    """
    for parser in context:
        if parser.ptype == ":Retrieve":
            break
        elif parser.ptype == ":Capture":
            p = cast(UnaryParser, parser).parser
            while p.ptype in (":Synonym", ":Forward"):
                p = cast(UnaryParser, p).parser
            if not isinstance(p, Retrieve):
                return True
    return None


862
863
864
865
866
867
########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################

868
869
def mixin_comment(whitespace: str, comment: str) -> str:
    """
870
    Returns a regular expression pattern that merges comment and whitespace
871
    regexps. Thus comments can occur wherever whitespace is allowed
872
873
874
875
876
877
    and will be skipped just as implicit whitespace.

    Note, that because this works on the level of regular expressions,
    nesting comments is not possible. It also makes it much harder to
    use directives inside comments (which isn't recommended, anyway).
    """
878
    if comment:
eckhart's avatar
eckhart committed
879
880
        whitespace = '(?:' + whitespace + ')'
        comment = '(?:' + comment + ')'
881
882
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
883
884


di68kap's avatar
di68kap committed
885
def mixin_nonempty(whitespace: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
886
    r"""
887
888
889
    Returns a regular expression pattern that matches only if the regular
    expression pattern `whitespace` matches AND if the match is not empty.

di68kap's avatar
di68kap committed
890
891
    If `whitespace`  does not match the empty string '', anyway,
    then it will be returned unaltered.
892

eckhart's avatar
eckhart committed
893
    WARNING: `minin_nonempty` does not work for regular expressions the matched
894
895
896
897
    strings of which can be followed by a symbol that can also occur at
    the start of the regular expression.

    In particular, it does not work for fixed size regular expressions,
eckhart's avatar
eckhart committed
898
899
900
    that is / / or /   / or /\t/ won't work, but / */ or /\s*/ or /\s+/
    do work. There is no test for this. Fixed-size regular expressions
    run through `mixin_nonempty` will not match at any more if they are applied
Eckhart Arnold's avatar
Eckhart Arnold committed
901
902
    to the beginning or the middle of a sequence of whitespaces!

eckhart's avatar
eckhart committed
903
    In order to be safe, your whitespace regular expressions should follow
Eckhart Arnold's avatar
Eckhart Arnold committed
904
905
    the rule: "Whitespace cannot be followed by whitespace" or "Either
    grab it all or leave it all".
906
907
908
909
910
911
912
913
914
915

    :param whitespace: a regular expression pattern
    :return: new regular expression pattern that does not match the empty
        string '' any more.
    """
    if re.match(whitespace, ''):
        return r'(?:(?=(.|\n))' + whitespace + r'(?!\1))'
    return whitespace


916
917
918
919
920
# AnalysisError = Tuple[str, Parser, Error]      # pname, parser, error
class AnalysisError(NamedTuple):
    symbol: str
    parser: Parser
    error: Error
921
922
923
924
925
926


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
927
    def __init__(self, static_analysis_result: List[AnalysisError]):
928
929
930
931
932
933
934
935
936
937
        assert static_analysis_result  # must not be empty
        self.errors = static_analysis_result

    def __str__(self):
        if len(self.errors) == 1:
            return str(self.errors[0][2])
        return '\n' + '\n'.join(("%i. " % (i + 1) + str(err_tuple[2]))
                                for i, err_tuple in enumerate(self.errors))


938
939
940
941
942
943
944
945
946
947
948
949
950
class Grammar:
    r"""
    Class Grammar directs the parsing process and stores global state
    information of the parsers, i.e. state information that is shared
    accross parsers.

    Grammars are basically collections of parser objects, which are
    connected to an instance object of class Grammar. There exist two
    ways of connecting parsers to grammar objects: Either by passing
    the root parser object to the constructor of a Grammar object
    ("direct instantiation"), or by assigning the root parser to the
    class variable "root__" of a descendant class of class Grammar.

eckhart's avatar
eckhart committed
951
    Example for direct instantiation of a grammar::
952

953
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
954
955
956
957
958
959
960
961
962
963
964
965
966
967
        >>> number_parser = Grammar(number)
        >>> number_parser("3.1416").content
        '3.1416'

    Collecting the parsers that define a grammar in a descendant class of
    class Grammar and assigning the named parsers to class variables
    rather than global variables has several advantages:

    1. It keeps the namespace clean.

    2. The parser names of named parsers do not need to be passed to the
       constructor of the Parser object explicitly, but it suffices to
       assign them to class variables, which results in better
       readability of the Python code.
968
       See classmethod `Grammar._assign_parser_names__()`
969
970

    3. The parsers in the class do not necessarily need to be connected
971
972
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
973
974
975
976
977
978
979

    As a consequence, though, it is highly recommended that a Grammar
    class should not define any other variables or methods with names
    that are legal parser names. A name ending with a double
    underscore '__' is *not* a legal parser name and can safely be
    used.

980
    Example::
981
982
983
984
985
986
987
988
989

        class Arithmetic(Grammar):
            # special fields for implicit whitespace and comment configuration
            COMMENT__ = r'#.*(?:\n|$)'  # Python style comments
            wspR__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)

            # parsers
            expression = Forward()
            INTEGER = RE('\\d+')
990
991
992
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
993
994
995
996
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
997
998
999
1000
    For any parser that is directly assigned to a class variable the
    field `parser.pname` contains the variable name after instantiation
    of the Grammar class. The parser will never the less remain anonymous
    with respect to the tag names of the nodes it generates, if its name
1001
    is matched by the `disposable__` regular expression.
1002
    If one and the same parser is assigned to several class variables
1003
    such as, for example, the parser `expression` in the example above,
1004
    which is also assigned to `root__`, the first name sticks.
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016

    Grammar objects are callable. Calling a grammar object with a UTF-8
    encoded document, initiates the parsing of the document with the
    root parser. The return value is the concrete syntax tree. Grammar
    objects can be reused (i.e. called again) after parsing. Thus, it
    is not necessary to instantiate more than one Grammar object per
    thread.

    Grammar classes contain a few special class fields for implicit
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:

1017
    Class Attributes:
1018
1019
1020
1021
        root__:  The root parser of the grammar. Theoretically, all parsers of the
                 grammar should be reachable by the root parser. However, for testing
                 of yet incomplete grammars class Grammar does not assume that this
                 is the case.
1022